DILAB | Data Intelligence Lab

Research Areas

System-based outcome evaluation and causality-based behavior assessment for model robustness and explainability

Knowledge-enhanced generation techniques and foundation model development

Multimodal large language models and their applications

Overthinking Loops in Agents: A Structural Risk via MCP Tools

arXiv 2026

TextME: Bridging Unseen Modalities Through Text Descriptions

ICML 2026

Towards Fully-Automated Materials Discovery via Large-Scale Synthesis Dataset and Expert-Level LLM-as-a-Judge

CIKM 2025

Sep 2026 One paper got accepted in WMT 2026, co-located with EMNLP 2026.

HUFS-DILAB at WMT 2026 Automated Translation Quality Evaluation Task 1: Language-Pair Prompt Routing for Error Span Annotation

Aug 2026 Selected for the NRF Core Research Program (기본연구B), 2026–2029.

Multi-faceted and Efficient Diagnosis of Large Language Model Capabilities (거대언어모델 능력의 다면적·효율적 진단 방법 연구)

May 2026 One paper got accepted in ICML 2026.

TextME: Bridging Unseen Modalities Through Text Descriptions

Feb 2026 New preprint released on arXiv.

Overthinking Loops in Agents: A Structural Risk via MCP Tools

Jan 2026 DILAB Homepage Created

Welcome to the new Data Intelligence Lab homepage.

Aug 2025 Two papers got accepted in CIKM 2025.

Towards Fully-Automated Materials Discovery via Large-Scale Synthesis Dataset and Expert-Level LLM-as-a-Judge

Rethinking the Training Paradigm of Discrete Token-Based Multimodal LLMs: An Analysis of Text-Centric Bias

Jun 2025 One paper got accepted in TACL and will be presented at EMNLP 2025.

Overcoming Source Object Grounding for Semantic Image Editing